Post Thumbnail

Hugging Face challenges DeepSeek: Project Open-R1 reveals secrets of Chinese AI

Hugging Face challenges DeepSeek: Project Open-R1 reveals secrets of Chinese AI The Hugging Face team presented the first results of the Open-R1 project aimed at reproducing the technologies of Chinese artificial intelligence DeepSeek-R1. Within a week, researchers managed to achieve significant progress in understanding and replicating this advanced system.

A key achievement was the successful reproduction of test results on the MATH-500 benchmark. Researchers confirmed impressive performance of various model versions: DeepSeek-R1-Distill-Qwen-32B achieved 95.0% accuracy compared to the claimed 94.3%, while the Llama-70B-based version showed 93.4% versus the official 94.5%.

During the study, a unique feature of DeepSeek-R1 was discovered – unprecedented length of generated responses. Analysis of distribution in the OpenThoughts dataset showed that the average response length is about 6000 tokens, and in some cases exceeds 20,000 tokens. “Considering that an average page contains approximately 500 words, and one token is slightly shorter than a word, many responses exceed 10 pages in volume,” researchers note.

To ensure research transparency, the Hugging Face team created an open Open-R1 leaderboard where the community can track progress in reproducing results. Special attention is paid to the issue of significant GPU memory requirements during training due to the need to generate long sequences.

The Open-R1 project, launched just a week ago, combined the efforts of various teams and the developer community. The main goal remains to reproduce the training pipeline and synthetic data of DeepSeek-R1, which will help better understand the operating principles of this advanced artificial intelligence system.

This initiative demonstrates a growing trend towards openness and collaboration in AI, where even the most complex technological achievements become the subject of collective study and reproduction by the global developer community.

Autor: AIvengo
For 5 years I have been working with machine learning and artificial intelligence. And this field never ceases to amaze, inspire and interest me.
Latest News
Amazon turns couriers into cyborgs with AI smart glasses

Amazon decided to turn its couriers into cyborgs. No, seriously - the company announced smart glasses with AI for delivery workers. The idea, according to the e-commerce giant, is to free up drivers' hands. And spare them from constantly switching gaze between phone, package and surroundings.

OpenAI will add character cameos to Sora

OpenAI published the development roadmap for Sora, and you know what? It seems the company finally realized that video generation isn't just a technological demonstration. But a tool that people need to actually use. Bill Peebles, project head, announced a whole set of updates, and some of them are really interesting.

Starcloud launches AI satellite in November for data center in space

You know where data centers for AI are now being moved? To space. Startup Starcloud, participant in NVIDIA Inception program, plans to launch in November a satellite with AI into Earth orbit. And this is only the beginning of their ambitious plan to solve problems of energy consumption and cooling of data centers on Earth.

ChatGPT Atlas is vulnerable to prompt injections and can help with phishing

I told about how OpenAI released the ChatGPT Atlas browser. And here the first users already found a whole bouquet of problems. Let's start with basic things. The browser has no built-in ad blocker, reading mode and text translation function on the page. To retell an article or translate it, you need to ask the bot in chat.

China broke ASML lithography machine while trying to copy it

Here's a story about how import substitution faced harsh reality. China tried to disassemble an ASML lithography machine of the DUV category to study its design. Result? They damaged the system and turned to ASML itself with a request to repair. The absurdity of the situation is off the charts.